estimating abundance of the remnant apennine brown bear population using multiple noninvasive...

206

Journal of Mammalogy, 96(1):206–220, 2015DOI:10.1093/jmamma/gyu029

Estimating abundance of the remnant Apennine brown bear population using multiple noninvasive genetic data sources

P. Ciucci*, V. Gervasi, L. Boitani, J. Boulanger, D. Paetkau, R. Prive, and E. Tosoni

Department of Biology and Biotechnologies, La Sapienza University of Rome, Viale dell’Università 32, 00185 Rome, Italy (PC, VG, LB, ET)Integrated Ecological Research, 924 Innes Street, Nelson, British Columbia V1L 5T2, Canada (JB)Wildlife Genetics International, Suite 200, 182 Baker Street, P.O. Box 274, Nelson, British Columbia V1L 5P9, Canada (DP, RP)

* Correspondent: [email protected]

Accurate and precise estimates of population size are critical for effective management but can be particularly difficult to achieve for small populations of large carnivores. We approached this challenge by integrating multiple noninvasive data sources into a DNA-based mark–recapture framework to estimate the abundance of the small and endangered Apennine brown bear population. To improve sample size and coverage, we collected hair samples from June to September 2011 by concurrently using 4 noninvasive sampling methods: intensive hair-snagging (forty-three 5 × 5-km cells and five 12-day sampling sessions) plus secondary sampling methods (bear rub trees, alpine buckthorn aggregations, and incidental sampling). Following marker selection based on tissue samples from 55 Apennine bears, we used 13 microsatellites (plus gender) and quality assurance protocols to identify multilocus genotypes from hair samples. We used Huggins closed models in program MARK to estimate population size, which allowed us to account for spatial, temporal, and demographic components of heterogeneity in secondary sampling methods. Based on 529 analyzed hair samples, 80.5% of which yielded high-confidence scores for all markers, we achieved a rather precise (CV = 7.9%) population estimate of 51 bears (95% CI = 47–66) including cubs. Compared to a previous survey in 2008, our results provide evidence that the Apennine brown bear population has not been declining in recent years. Additionally, the relatively high (closure corrected) density (39.7 bears/1,000 km2; 95% CI = 36.6–51.4) indicates that habitat productivity within the core range is currently adequate for bears and that effective conservation of this small bear population should aim to expand the bears’ range across a larger portion of the central Apennines. We examined if a reduction in sampling effort would affect the precision of our population estimates. Reduced sample coverage, small sample size, and low hair-trap-capture probability preclude the adoption of a single sampling method or a subset of such to survey small bear populations if a comparable level of precision is required.

Key words: Apennine brown bear, hair-snagging, Huggins model, mark–recapture, noninvasive genetic sampling, population size, program MARK, small populations, Ursus

© 2015 American Society of Mammalogists, www.mammalogy.org

Small populations of large carnivores pose the greatest theo-retical and practical challenges for population assessment and monitoring (Clevenger and Purroy 1996; McDonald 2004; Jackson et al. 2006; Lynam et al. 2009). Reduced population sizes of rare and elusive carnivores likely correspond to exceed-ingly small and sparse data sets for reliable estimation of abun-dance (Janecka et al. 2008; Lynam et al. 2009; Gervasi et al. 2010; Gerber et al. 2014). Increased uncertainty of population abundance, in turn, reduces the power of monitoring programs to detect real changes in the population (McComb et al. 2010), undermining the timely adoption of adaptive management (Nichols and Williams 2006). However, the recent availability

of a variety of noninvasive sampling methods for carnivores (e.g., Long et al. 2008; O’Connell et al. 2011; Kelly et al. 2012) coupled with the development of flexible modeling approaches are increasingly offering new opportunities to reliably assess the status of small populations (e.g., Cubaynes et al. 2010; De Barba et al. 2010; Gerber et al. 2014).

Noninvasive genetic sampling (Schwartz et al. 2007) is increasingly being integrated into mark–recapture frame-works to estimate the abundance of carnivore populations (e.g., Boulanger et al. 2004; Mulders et al. 2007; Mondol et al. 2009; Ruell et al. 2009). Concurrently, modern mark–recapture mod-eling allows individual, environmental, or group covariates to

mailto:[email protected]?subject=

http://www.mammalogy.org

CIUCCI ET AL.—SMALL BEAR POPULATION SIZE ESTIMATION 207

account for heterogeneity in capture probability more efficiently (White et al. 2002; Chao and Huggins 2005). In addition, the simultaneous adoption of multiple data sources has been shown to improve the accuracy and precision of population size esti-mates (Boulanger et al. 2008; Sollmann et al. 2013).

Pioneered by Woods et al. (1999), noninvasive genetic sam-pling has become the established practice for estimating bear population abundance in North America (e.g., Mowat and Strobeck 2000; Boulanger et al. 2002; Kendall et al. 2008, 2009; Proctor et al. 2010; Sawaya et al. 2012), even though the application of this technique as a stand-alone sampling method is expected to produce sparse data sets in small bear popula-tions (Proctor et al. 2005, 2010; Gervasi et al. 2008). However, using sampling methods that ensure sufficient sample coverage and high levels of capture probability increases the accuracy of abundance estimates (De Barba et al. 2010; Gervasi et al. 2010). Additionally, the accuracy of DNA-based surveys can be enhanced with quality assurance protocols to detect and eliminate genotyping errors (Waits and Paetkau 2005) and using a sufficient number of variable genetic markers to reli-ably identify individuals (Mills et al. 2000; Waits et al. 2001). Both aspects are especially relevant in the case of small and genetically depleted bear populations (Paetkau 2003).

Brown bears in the Apennines (Ursus arctos marsicanus), central Italy, survive in a relict and long-isolated population, representing 1 of 4 very small bear populations in southwestern Europe whose conservation status is critical (Zedrosser et al. 2001). Despite long-time protection afforded by national and regional authorities to Apennine bears, range expansion has not been observed in recent decades, probably due to persistently high levels of human-induced mortality (Ciucci and Boitani 2008). The first formal estimate of the size of the Apennine brown bear population was produced in 2008, indicating 40 (95% CI = 37–52) bears in the core range (Gervasi et al. 2012). The action plan for the conservation of the Apennine brown bear (Anonymous 2011) emphasized the urgent need to produce reliable estimates of the bear population size to assess its trends and measure the outcome of conservation actions. Accordingly, we started such an effort in 2011. Building upon our previous experience in surveying this small bear population (Gervasi et al. 2008, 2012), we expected that hair-snagging data alone would be too sparse to yield an accurate and precise population estimate. The 2008 survey tackled this problem by integrat-ing multiple data sources into a DNA-based mark–recapture framework (Boulanger et al. 2008). We took advantage of pre-viously handled bears in the population and integrated data from hair-snagging, live captures, and direct observations of marked bears, including females with cubs, to obtain a rather precise population size estimate (CV = 6.8%—Gervasi et al. 2012). This approach, however, entailed costly survey condi-tions (i.e., availability of marked bears) and required a particu-larly high capture probability to overcome the potential bias due to correlation among data sources (Gervasi et al. 2012), making it impractical in the long term. Additionally, although the estimated match probabilities (i.e., PID and PIDSIB—Paetkau 2003) indicated the marker system we used was adequate for

individual identification, a thorough screening of the larger set of markers used in other bear populations (Taberlet et al. 1997; Paetkau et al. 1998; Paetkau 2003) is still lacking for the Apennine brown bear population.

We report here on the noninvasive genetic survey of the Apennine brown bear population that we conducted in 2011. Our aims were 3-fold. First, we formally assessed an array of microsatellites to determine an ideal set of markers for indi-vidual identification in this genetically depleted bear popula-tion. Second, we wanted a precise estimate of the size of the Apennine brown bear population based solely on noninvasive sampling. To accomplish this objective, we combined 4 non-invasive genetic sampling methods to develop encounter his-tories for use in mark–recapture closed-population models (Boulanger et al. 2008). Third, with the specific aim of deter-mining a cost-effective sampling design for future DNA-based surveys of this bear population, we explored the effects of pro-gressively reducing the combination of sampling methods on the abundance estimates through comparison with the estimate obtained using the full data set.

Materials and MethodsStudy area.—The 1,221-km2 sampling grid largely over-lapped with the National Park of Abruzzo, Lazio and Molise (PNALM), its external buffer zone and adjacent areas of known bear presence (PNALM ecosystem; Fig. 1). Corresponding to

Fig. 1.—Location of hair traps distributed within the forty-three 5 × 5 km cells used to survey the Apennine brown bear (Ursus arctos mar-sicanus) population in the PNALM ecosystem, Italy (inset), from June to September 2011. Location of rub trees and buckthorn (Rhamnus alpinus) sites installed with hair traps is also shown.

https://www.researchgate.net/publication/5361093_Multiple_Data_Sources_Improve_DNA-Based_Mark-Recapture_Population_Estimates_of_Grizzly_Bears?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==



https://www.researchgate.net/publication/235662102_A_Preliminary_Estimate_of_The_Apennine_Brown_Bear_Population_Size_Based_on_Hair-Snag_Sampling_and_Multiple_Data_Source_Mark-Recapture_Huggins_Models?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

https://www.researchgate.net/publication/235662088_A_multiple_data_source_approach_to_improve_abundance_estimates_of_small_populations_The_brown_bear_in_the_Apennines_Italy?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==


https://www.researchgate.net/publication/7522322_Genetic_analysis_reveals_demographic_fragmentation_of_grizzly_bears_yielding_vulnerably_small_populations_Proc_R_Soc_B?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

https://www.researchgate.net/publication/264610284_Using_multiple_data_sources_provides_density_estimates_for_endangered_Florida_panther?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

https://www.researchgate.net/publication/232685629_Noninvasive_genetic_sampling_tools_for_wildlife_biologists_A_review_of_applications_and_recommendations_for_accurate_data_collection?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

208 JOURNAL OF MAMMALOGY

the core distribution of the Apennine brown bear population (Ciucci and Boitani 2008), the PNALM ecosystem is located across the central Apennines, with the main mountain ridges extending NW–SE and elevation ranging from 400 to 2,285 m. Mean monthly temperatures range from 2°C in January to 20°C in July, with rainfall most frequent in spring and fall and snow cover generally extending from mid-December to March. Deciduous forests cover about 60% of the study area, com-posed primarily of beech (Fagus sylvatica) but also including oaks (Quercus cerris and Q. pubescens) at lower altitudes. Sub-alpine meadows and grasslands cover approximately 22% of the study area (EEA 2006). Road density is 1.1 km/km2, includ-ing unpaved roads, and human density averages 14.6 inhabit-ants/km2. Forest cutting in the PNALM is strictly regulated by the Park authority and hunting is prohibited within the PNALM although hunting with dogs of wild boar, hare, and birds is allowed in the external buffer area. Additional details of the study area can be found elsewhere (Ciucci and Boitani 2008; Falcucci et al. 2009; Ciucci et al. 2014).

Prior to this study, 3 noninvasive sampling applications had been conducted on this bear population including a preliminary survey in 2004 (Gervasi et al. 2008), a restricted pilot study

in 2007 (Gervasi et al. 2010), and a formal population esti-mation in 2008 (Gervasi et al. 2012). A minimum of 50 bear genotypes were detected in these surveys, comprising dead and living bears at the time of this study. Due to topographic, land-scape, and anthropogenic differences which sharply demarcate the northern, western, and southern borders of the study area, geographic closure represents a minor concern in surveying our bear population (Gervasi et al. 2012).

Sampling.—We concurrently adopted 4 noninvasive sam-pling methods to collect bear hair for genetic analysis. We used traditional hair-trap sampling (Wood et al. 1999) as a primary sampling method. To systematically distribute hair-trapping effort throughout the study area (Kendall et al. 2008, 2009; Proctor et al. 2010), we used a sampling grid of forty-three 5 × 5 km cells (Fig. 1). Within each cell, preliminary onsite inspec-tions in high-quality habitats indicated ideal trap locations to enhance capture probability (Gervasi et al. 2012). We used five 12-day sampling sessions (1 June–26 July 2011; Table 1) for a total of 215 traps. Each trap consisted of a 25–30 m perimeter of barbed wire around 4–7 large trees at a height of 50 cm (Woods et al. 1999). We used about 5–6 liters of a 2:1 mixture of aged cattle blood and decomposed fish oil as a scent lure, pouring it

Table 1.—Results of 4 noninvasive sampling methods used to survey the Apennine bear (Ursus arctos marsicanus) population in its core dis-tribution (PNALM, June–July 2011).

Sampling method Date Sessiona Number of trapsb Sampling effortc Successful trapsd Total no. of bear samples

Bear samples/trapb

X (± SD) Range

Hair-snagginge 1–12 Jun. 1 43 — 10 (23%) 76 7.6 (± 7.6) 2–2713–24 Jun. 2 43 — 5 (12%) 32 6.4 (± 5.8) 1–16

25 Jun.–6 Jul. 3 43 — 7 (16%) 24 3.4 (± 1.9) 1–77–18 Jul. 4 43 — 6 (14%) 15 2.5 (± 1.0) 1–419–30 Jul. 5 43 — 6 (14%) 12 2.0 (± 1.1) 1–4

Buckthorn samplingf

17–25 Aug. — 6 2,432 0 0 — —26 Aug.–3 Sep. 6 7 4,532 4 (57%) 52 13.0 (± 12) 1–30

4–12 Sep. 7 7 5,409 6 (86%) 70 11.7 (± 8.6) 2–2613–21 Sep. 8 7 5,409 3 (43%) 17 5.7 (± 6.4) 1–1322–30 Sep. — 7 2,915 0 0 — —

Rub-tree samplingg

8–19 Jun. 9 56 65 8 (14%) 26 3.2 (± 1.6) 1–620 Jun.–1 Jul. 10 88 290 19 (22%) 41 2.2 (± 1.4) 1–7

2–13 Jul. 11 66 302 16 (24%) 38 2.2 (± 1.3) 1–614–25 Jul. 12 64 518 24 (37%) 60 2.5 (± 1.5) 1–7

26 Jul.–6 Aug. 13 75 574 16 (21%) 34 2.1 (± 0.9) 1–47–18 Aug. 14 83 304 10 (12%) 23 2.3 (± 1.3) 1–519–30 Aug. 15 46 93 3 (6%) 5 1.7 (± 0.5) 1–2

31 Aug.–11 Sep. 16 58 395 8 (14%) 14 1.8 (± 0.4) 1–212–23 Sep. 17 56 359 8 (14%) 11 1.4 (± 0.5) 1–2

24 Sep.–5 Oct. 18 18 825 16 (89%) 26 2.1 (± 0.7) 1–3Incidental sampling

1 Jun.–30 Sep. 19 — — — 67 — —

a Sessions are sequentially numbered as for modeling purposes. The first and last buckthorn sampling sessions were excluded from modeling as no samples were collected.b Either buckthorn sites (buckthorn sampling) or rub trees (rub-tree sampling).c Where relevant for modeling. Rub-tree sampling: cumulative number of days since the last visit at each active rub tree. Buckthorn sampling: total number of trap nights per session

multiplied by length of barbed wire installed.d In parenthesis: percentage of successful traps, or buckthorn sites, or rub trees.e The average length of each sampling session (12 days) might have varied ± 2 days because the date of deactivation of the trap could have been anticipated or postponed by 1–2 days.f Session dates depend on installation date and vary slightly by buckthorn site.g Sessions were defined according to 12-day intervals for modeling purposes.

https://www.researchgate.net/publication/250146828_The_Apennine_Brown_Bear_A_Critical_Review_of_Its_Status_and_Conservation_Problems?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==






on a centrally placed pile of rotten wood covered with leaves, moss, and other forest debris (Woods et al. 1999; Kendall et al. 2008). To enhance capture probability and reduce the risk of behavioral responses, we moved traps (≥ 1 km) within each cell at the end of each 12-day session. The borders of peripheral cells were drawn along ridges, valley bottoms, or other relevant landscape features (Mowat and Strobeck 2000; Proctor et al. 2010) to further improve the geographic closure of the sam-pling grid (Gervasi et al. 2012). We checked hair traps at the end of each session, collected hair samples, and dismantled and moved the traps to a new location.

We did not expect to detect cubs in our brown bear popu-lation, as they are inaccessible to 50-cm hair traps in spring (Gervasi et al. 2012). While we could have added a 2nd lower strand on all hair traps, this would have increased logistic com-plexity to unsustainable levels, so we used double-stranded hair traps in buckthorn traps only (see below). As a second-ary sampling method, we inventoried a minimum of 19 rub trees/100 km2 and installed hair traps on 97 of them from June to September 2011 (Table 1). We attached 4–6 strands of barbed wire (30–40 cm each) to the rubbing surface of the tree in a zig-zag pattern between approximately 30 and 170 cm from the tree base (Kendall et al. 2008) and recorded the rub-tree location with a GPS. As bear rubbing is a naturally occurring behavior (Green and Mattson 2003), we did not use lures or other bear attractants. Effective sampling ranged from 53 to 120 days per rub tree (X SD= ±1 7 14 0 ), during which we visited installed rub trees at 1- to 13-day intervals, for a total of 833 visits ( X SD= ±9 4 visits per rub tree). We only collected hair samples from the barbed wire and not from other rubbing sur-faces to be sure of the time of deposition (Kendall et al. 2008). Hair traps on rub trees were not simultaneously installed or dismantled, and so collected samples were organized into ten 12-day sessions (Table 1).

We also adopted opportunistic sampling at buckthorn patches (Rhamnus alpinus), a particularly effective sampling method in our bear population (Gervasi et al. 2008). We sampled bears at 7 buckthorn sites across the PNALM (Fig. 1), during the rip-ening period of buckthorn berries (17 August–30 September 2011; Table 1). At each buckthorn site, we constructed 1–4 peripheral hair traps by encircling cohesive buckthorn aggrega-tions with double-stranded barbed wire at 30 and 50 cm high to make cubs accessible to sampling. Cumulative length of hair traps at each buckthorn site ranged 65–130 m. As foraging on buckthorn berries is a naturally occurring behavior (Ciucci et al. 2014), we did not use any attractants and did not expect a behavioral response to sampling. We checked each buckthorn trap 4–5 times during the sampling period at a mean interval of 8 (± 2) days and dismantled all traps simultaneously at the end of the Rhamnus ripening season. Due to differences in installation dates, the sampling period ranged from 31 to 39 ( X SD= ±35 3 ) days per buckthorn site. We organized col-lected samples into five 8-day sessions.

Incidental sampling, from June to September 2011, was the final secondary sampling method incorporated in our survey. This included bear hair collected by park wardens during their

patrolling activity and assessment of property damage caused by bears (Gervasi et al. 2008). A hair sample was defined as a tuft of guard hairs with bulbs entangled in 1 set of barbs (Woods et al. 1999) for all sampling methods. We collected each sample using sterilized surgical forceps and placed each sample in a paper envelope labeled with a uniquely numbered barcode (Kendall et al. 2009). We then passed a flame under the barbs to remove any trace of hair to avoid contamination between successive sessions. Collected samples were stored in a box with silica gel in a dark place to avoid DNA degrada-tion. We discarded all obvious nonbear samples in the field and used microscopic characteristics (Teerink 1991) and an expe-rienced observer (ET) to further distinguish dubious samples. Therefore, only macro- and microscopically preselected bear samples were considered for genetic analysis.

Genetic methods.—To identify individual genotypes from hair samples, we used a laboratory specializing in the analysis of noninvasive genetic samples (Wildlife Genetics International, Nelson, British Columbia). Blood samples drawn from handled bears during 2005–2009 (Gervasi et al. 2012—L. Gentile, Abruzzo National Park, pers. comm.) were provided in soaked dried cotton swabs, and DNA was extracted by processing a ~3 × 3 mm piece of cotton swab with QIAGEN DNeasy Blood and Tissue Kits (Qiagen, Valencia, California). We analyzed hair samples following the same procedure, washing them in warm water before placing them in the extraction solution. We aimed to use 10 clipped guard hair roots per sample or, alterna-tively, entire clumps of whole underfur.

We analyzed samples from 25 Apennine brown bears, com-prising dead and living individuals at the time of this study, for which we had blood samples from 2005 to 2009. We summa-rized variability of 30 microsatellites from these samples using GENEPOP (Raymond and Rousset 1995) to provide marker selection data (Table 2; Appendix I). When selecting markers, we had to retain a sufficient number of the microsatellites used in the previous laboratory to ensure a low-match probability in comparisons between current genotypes and those detected in previous surveys (Gervasi et al. 2008, 2012). To identify indi-viduals from hair samples in this study, we selected 13 mic-rosatellites, plus amelogenin for gender (Ennis and Gallagher 1994—see below), deferring the selection of an ideal set of markers for use in future studies until all hair samples were analyzed, allowing assessment of marker power to be based on a larger number of genotypes.

During a first pass, we analyzed all extracted hair samples at 7 of the 14 markers and culled samples that failed to pro-duce high-confidence scores (i.e., strong and typical genotype profiles—Kendall et al. 2009) for ≥ 4 of 7 markers. This was followed by a cleanup phase in which we reanalyzed data points that were scored with low confidence, using 5 µl of DNA per reaction instead of the 3 µl used during the first pass. Multiple rounds of reanalysis were used to confirm persistently weak data points. The first pass and cleanup process were then repeated at the other 7 markers with the nonculled hair samples, and further samples were eliminated after the cleanup phase of this 2nd round of 7-locus genotyping. Analyzing gender on






https://www.researchgate.net/publication/229531524_Demography_and_Genetic_Structure_of_a_Recovering_Grizzly_Bear_Population?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==


https://www.researchgate.net/publication/31011134_GENEPOP_Population_Genetics_Software_for_exact_tests_and_ecumenicism?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==


all samples, rather than limiting the analysis to 1 sample per individual after the microsatellite analysis was finalized (e.g., Sawaya et al. 2012), allowed us to roughly halve the match probability, even for close relatives. Samples left after this final cull had high-confidence scores for all 14 markers. The detec-tion of mixture is particularly challenging when low variability results in many markers having only 2 common alleles, since the most reliable indication of a mixed sample is the amplifica-tion of > 2 alleles at a given marker. For this reason, we were careful to withhold high-confidence scores in cases where peak heights were reproducibly imbalanced, as occurs when a mix-ture comprises 1 heterozygote and 1 homozygote.

Genotyping errors may bias mark–recapture estimates of population size (Mills et al. 2000; Roon et al. 2005). Thus, we replicated the mismatching markers in any pair of genotypes similar enough to have conceivably been created by genotyp-ing error (Paetkau 2003). Intensive testing with blind con-trol samples has shown that this protocol effectively prevents the recognition of false individuals through genotyping error (Kendall et al. 2009; Proctor et al. 2010). During error-check-ing, 6 errors were found and corrected. After correction, the most similar pair of genotypes in the data set mismatched at 2 markers (i.e., 2MM-pairs), and those mismatching data points had been solidly replicated to rule out genotyping error. Finally, we performed a series of cross-controls between genotypes and information from field data, aiming to verify locations of samples attributed to previously radiocollared bears with their general home range location, unexpectedly large distances

between samples attributed to the same genotype, and consis-tency of genotype sampling at each trap, evaluating collection date, the position of samples on the trap, and the number of mismatching loci among genotypes sampled at the same trap. Five suspicious cases emerged from this post quality control, which were reanalyzed, leading to 1 more potentially mixed sample being culled from the data set.

Once all hair samples were analyzed, we sought to identify an ideal set of markers for individual identification in this bear population. We used expected heterozygosity and the observed number of alleles at each locus to quantify genetic variation. Although probability of identity (PID and PID SIB—Paetkau 2003) is routinely used to describe the power of a given set of markers to differentiate individual multilocus genotypes (Paetkau and Strobeck 1998), match probabilities assume a specified level of relatedness making it difficult to interpret them in the context of a study population in which the distribution of consanguin-ity is unknown (Kendall et al. 2009; Proctor et al. 2010). This constraint is especially apparent in a study population such as ours, where consanguinity is expected to be high. We there-fore obtained a more direct empirical estimate of match prob-ability by extrapolating from observed mismatch distributions (Paetkau 2003; Kendall et al. 2009; Proctor et al. 2010). Such extrapolation is robust across a range of marker variability and degrees of consanguinity, because the distribution of degrees of consanguinity is implicit in the mismatch distribution, thus pro-viding a reasonable estimate of how many 0MM-pairs might be sampled with a given set of markers (Paetkau 2003). We sought evidence of cubs in our 2011 sample by examining newly detected genotypes that shared at least 1 allele with their puta-tive mother (limited to known adult females) at all 13 loci and with whom they were sampled in the same trap and sampling occasion. We further corroborated such cases by confirming the presence of females with cubs in the same general area by direct sightings. These criteria were meant to provide evidence of the accessibility of cubs to sampling, and not to identify all cubs in the sample, and were applicable to our bear population as most yearlings leave their mothers by June of their 2nd year (Tosoni 2010).

Data analysis.—We used Huggins closed-population models (Huggins 1991) in program MARK (White and Burnham 1999) to estimate the size of the Apennine brown bear population. We first combined the data from the 4 noninvasive sampling methods to construct individual encounter histories. For each sampled bear, we assigned hair-snag captures to sessions 1–5, captures at buckthorn sites to sessions 6–8, rub-tree captures to sessions 9–18, and incidental samples to session 19. This approach is feasible in a context of closed population capture–recapture models, as the relative order of sessions is irrelevant to parameter estimation, unless any behavioral response is expected in the data (Boulanger et al. 2008). We did not expect our sampling design to be affected by any strong behavioral response (see “Sampling”).

We constructed candidate models for each data source based on bear biology and insights from previous surveys (e.g., Boulanger et al. 2008; Gervasi et al. 2008, 2012; Kendall et al. 2008, 2009). In addition to sampling method, we used

Table 2.—Variability of the Apennine bear (Ursus arctos marsica-nus) at the 13 microsatellite markers used for individual multilocus genotyping in this study plus G10P and gender. Marker selection was originally conducted on tissue and blood samples (≤ 25 bears), whereas reported variability for the 13 selected markers was successively calcu-lated including hair samples collected during the 2011 survey (n = 30 bears). Measures of variability include the observed number of alleles (A) and expected (HE) and observed (HO) heterozygosity.

Locus n HE HO A

CXX20 55 0.62 0.67 3REN144A06 55 0.61 0.65 3G1Da 55 0.58 0.64 3Mu51a 55 0.57 0.47 3G10Ba 55 0.52 0.47 3G10Ca 55 0.50 0.56 3Mu59a 55 0.49 0.53 2MSUT-2 55 0.49 0.36 3G10X 55 0.47 0.35 2Mu05a 55 0.47 0.47 2G10La 55 0.44 0.51 2Mu50a 55 0.44 0.47 2Mu11a 55 0.44 0.38 2G10Pa,b 24 0.22 0.25 2

a Markers used by another lab in previous surveys of the Apennine brown bear population (Gervasi et al. 2008, 2012).b In addition to the 13 markers above, G10P was used at a later stage of analysis to better discriminate equivocal cases (i.e., 2MM- and 1MM-pairs) between genotypes detected in 2011 and those detected during previous surveys and scored by another lab.



https://www.researchgate.net/publication/265996086_Some_Practical_Aspects_of_a_Conditional_Likelihood_Approach_to_Capture_Experiments?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==




https://www.researchgate.net/publication/231877762_A_simulation_test_of_the_effectiveness_of_several_methods_for_error-checking_non-invasive_genetic_data?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

https://www.researchgate.net/publication/224919561_Estimating_Grizzly_and_Black_Bear_Population_Abundance_and_Trend_in_Banff_National_Park_Using_Noninvasive_Genetic_Sampling?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==

https://www.researchgate.net/publication/216771416_Program_MARK_Survival_Estimation_from_Populations_of_Marked_Animals?el=1_x_8&enrichId=rgreq-7270bf70-4cb2-492f-be9c-3145de24afb9&enrichSource=Y292ZXJQYWdlOzI3NTAyMDAxNjtBUzoyMTg1MDc3NDE1MzYyNTZAMTQyOTEwNzcxMzU3Ng==


individual, group, and temporal covariates to model heteroge-neity in capture probability. For each of the sampling methods, except incidental sampling, we tested for temporal variation by time (t; 1 parameter for each session) and trend (T) effects. For hair-snagging data, we tested if a previous livetrapping event (lt) was associated with a decreased capture probability (Boulanger et al. 2008; Kendall et al. 2008, 2009). We also tested if a hair-snag detection in previous surveys (prev.hs) affected capture probability, as this covariate in our bear population can be used as a proxy for 2 broad age classes (i.e., ≤ 4 or > 4 years). In fact, because in the 2007 and 2008 hair-snag surveys (Gervasi et al. 2010, 2012) we did not detect new bear genotypes compared to those detected in the 2004 survey (Gervasi et al. 2008), all the bears sampled in 2011 which had been previously detected had to be > 4 years old (i.e., they were born before 2007). To sup-port this assertion, we estimated the probability that some bears present in the population before 2007 went undetected in the 2007 and 2008 surveys and were still alive in 2011 (Appendix II). Finally, we tested if the capture probability derived from hair-snag data alone varied as a function of gender (sex).

As the effective length of each buckthorn sampling session differed slightly among buckthorn sites, we tested the effect of variation in sampling effort (bte) using the cumulative number of sampling days in each session multiplied by the cumulative length of the barbed wire at each buckthorn site, the latter rang-ing 65–130 m per site. In addition, as buckthorn berries tend to ripen later in the season in the northern part of the study area, we tested an interaction between time (t) and latitude (lat) of each bear’s mean capture location (i.e., the geometric center of all its detection points). We also tested the effect of the distance from each bear’s mean capture location to the nearest installed buckthorn site (dbt) as installed buckthorn sites were not evenly distributed across the study area. Finally, we included a gender effect (sex) and an effect of previous hair-snag sampling (prev.hs; see above).

We modeled temporal variability in rub-tree sampling effort (rte) as the cumulative number of sampling days for each rub tree summed across all rub trees within each session (Kendall et al. 2008). Due to the markedly uneven distribution of installed rub trees across the study area, we also tested for the effect of the number of installed rub trees available to each sampled bear (nrub). We first created a buffer, equivalent to the average seasonal home range (50 and 115 km2 for females and males, respectively—Tosoni 2010), around the bears’ mean capture locations. We then tallied the installed rub trees in each bear’s buffer, weighted by the cumulative number of sampling days within each session. We also included a gender effect, an effect of previous hair-snag experience (prev.hs; see above), and the interaction between the latter 2 covariates. Because it was not possible to estimate sampling effort for incidental samples, we summarized the data in a single session and only tested for a gender effect and for the effect of previous hair-snag history (prev.hs; see above).

We hypothesized that 2 additional forms of spatial hetero-geneity might have affected all sampling methods. First, as violation of geographic closure would cause a decrease in

capture probability toward the periphery of the sampling grid (Boulanger and McLellan 2001; Boulanger et al. 2008), we tested the distance from each bear’s mean capture location to the edge of the grid (dfe). We also used log(dfe) and dfe2 to assess if different shapes of the relationship were more sup-ported by the data. Second, as our efficiency in selecting ideal locations for hair traps might be lower toward the more periph-eral and less known portions of the study area, we tested for an effect of the distance between each bear’s mean capture location and the NW–SE central mountain ridge of the study area (dfc). We tested for this effect both additively on all data sources and separately for each sampling method.

After generating the global model, we fitted reduced models and assessed their relative support using the Akaike informa-tion criterion adjusted for small sample size (AICc—Burnham and Anderson 2002). Similar to stepwise regression procedures using AIC to evaluate the relative support of nested models (Burnham and Anderson 2002), we started from a fully param-eterized model, which included all the potential effects, and eliminated them one by one, retaining the ones whose elimina-tion produced a marked increase in the AIC. In doing so, we fitted less-parameterized models for one sampling method at a time while keeping constant the structure of the model for the rest of the design. Once identified, the most parsimonious parameterization for a given sampling method was kept constant for the rest of the model selection procedure. We then repeat-edly fitted less-parameterized models for each of the remain-ing sampling methods until a final most parsimonious general model was identified. Compared to separately developing most parsimonious models for each data source and then combining them into a final model, our approach estimates the variation in capture probability in each data type while taking into account also those bears that, even if undetected by a given sampling method, have been detected by one or more of the others.

Estimates of the female, male, and total population size were obtained as derived parameters from the Huggins model, and we used model averaging for parameter estimation to account for model uncertainty (Burnham and Anderson 2002). We cal-culated 95% log-based confidence intervals of model-averaged population size accounting for the minimum number of bears known to be alive and in the study area during the survey period (Mt+1—White et al. 2002). The estimate derived from this pro-cedure refers to the superpopulation (sensu Kendall 1999), including bears residing partially or totally in the study area. We did not deem it necessary to correct such an estimate given the relatively isolated nature of the Apennine bear core distribu-tion (Gervasi et al. 2012). However, with reference to the sam-pling area, we corrected bear density for closure violation using a regression-based method (Ivan et al. 2013) implemented in program MARK and according to the average fidelity (95.1%) previously estimated in our study area based on GPS-collared bears (Gervasi et al. 2012). Compared to previous procedures to estimate fidelity (White and Shenk 2001), this method is less sensitive to the assumption that radiocollared bears have the same spatial distribution as the DNA bears, thus providing unbiased estimates of bear fidelity (and hence of density) when



livetrapping and radiotracking effort are not evenly distributed across the study area. Unsuitable areas (e.g., lakes, infrastruc-tures) represent a negligible portion of the surveyed area, and we did not exclude them from density calculation.

To provide practical insights into how to improve sampling-design efficiency, we explored the effects of progressively reduced sampling designs on magnitude and precision of abun-dance estimates compared to that obtained from the fully com-bined data set.

ResultsMarker selection and power.—Nine of the 30 microsatellites that we investigated were invariable, and another 7 ampli-fied just 2 alleles and showed HE < 0.35 (Appendix I). These biallelic markers included G10P (HE = 0.22) and MU15 (HE = 0.04), which were among the 11 microsatellites used in previous surveys. In addition to these 16 comparatively invari-able markers, we also excluded G10H (HE = 0.46) because we expected that the comparatively long alleles at this marker (> 250 bp) would compromise success rates. This left 13 mic-rosatellites (plus gender) for individual identification. Using this set of 14 markers, we identified 55 unique genotypes (see below), comprising all dead and extant bears sampled, with just 1 2MM- and no 1MM-pairs, and thus no realistic chance of having sampled any 0MM-pairs (Fig. 2a). This suggests that we used more markers than strictly necessary to achieve a low-match probability. However, as we needed to integrate our genotypes with those previously scored by another lab (Gervasi et al. 2008, 2012), we also evaluated the risk of false matches among the 55 genotypes using the 10 markers with HE > 0.22 (including gender) common to both labs (Table 2). This data set included ten 2MM-pairs and one 1MM-pair, suggesting a small but nontrivial risk of false matches (Fig. 2b). This risk further increased when we removed MU11, the least variable of the common microsatellites (Appendix I), indicating a real possibility of encountering false matches in any comparison limited to 9 markers (Fig. 2b). We therefore used 10 common markers in our routine analysis but added an 11th common marker (G10P) to any genotype that matched to, or was in a 1MM- or 2MM-pair with, any genotype scored by the previous lab, functionally extending the comparison between new and old data sets to 11 common markers (including gender). We then explored mismatch distributions by progressively adding the 4 other microsatellites used in our study to the 10 common markers. We found a single 2MM-pair and no 1MM-pairs with 12 markers, after removing G10X and MSUT-2, the least vari-able of the 4 markers that we added to the common set of 10 (HE < 0.50; Fig. 2a). Using data from 55 dead and living bears, mean observed heterozygosity across the 13 markers used in this study was 0.50 (± 0.03 SE), with an average of 2.54 (± 0.14 SE) alleles per locus (Table 2).

Sampling.—We collected 679 hair samples between 1 June and 30 September 2011 (Table 1). Hair-snagging yielded 159 bear hair samples from 34 (15.8%) hair traps in 25 (58.1%) sampling cells, with an average of 4.4 (± 2.5 SD) bear hair samples per trap. We collected 278 bear hair samples during

134 (16.1%) successful visits to 56 (57.7%) rub trees, yielding an average of 2.2 (± 0.7 SD) hair samples per rub tree per ses-sion. Buckthorn sampling yielded 139 hair samples collected at 6 out of 7 installed buckthorn sites although the first and the last sampling sessions provided no samples. Finally, incidental sampling yielded 67 hair samples.

Genetic analyses.—We used 599 hair samples for genetic analysis, 70 of which were not analyzed either because they contained no guard hair roots and < 5 underfur or because they were believed to be replicated samples (Table 3). Of 529 analyzed samples, 426 yielded high-confidence scores for all 13 microsatellites plus gender, for a success rate of 80.5% (Table 3). Proportionally more samples were culled for hair-snagging and incidental sampling than for rub-tree and buck-thorn sampling (Gadj = 19.8, d.f. = 3, P < 0.001). We detected 45 bears, comprising 15 previously livetrapped for research

Fig. 2.—Mismatch distributions for 55 Apennine brown bears (Ursus arctos marsicanus) based on a) an increasingly smaller subset of the group of 14 markers (13 microsatellites plus gender, cf. Table 2), showing that the removal of either or both G10X and MSUT-2 had lit-tle impact on match probability; the further removal of MU11 caused an increase in the number of 2MM-pairs; b) 10 markers scored by both Wildlife Genetics International and the genetic laboratory used in a previous study (Table 2; Appendix I).


purposes, 14 noninvasively sampled in previous surveys, and 16 newly detected bears. Three bears (2.2%) were detected from 1 sample only, 17 (37.8%) from 2 to 4 samples, 11 (24.4%) from 5 to 10 samples, 8 (17.8%) from 11 to 20 samples, and 6 (13.3%) from 22 to 71 samples. No single sampling method detected all bears: 3 bears (6.7%) were detected by all sampling methods, 5 (11.1%) by any combination of 3 sampling meth-ods, 17 (37.8%) by any combination of 2 sampling methods, and 20 (44.4%) by 1 sampling method only.

Hair-snagging detected the largest number of bears, fol-lowed by buckthorn, rub-tree, and incidental sampling (Table 3). Hair-snagging and buckthorn sampling contrib-uted the largest number of uniquely detected bears (8 bears each), cumulatively increasing by 80% the minimum num-ber of bears detected by the other 2 sampling methods. Hair-snagging and buckthorn sampling also contributed the largest number of bears not known from previous surveys (8 bears each; Table 3). Rub-tree sampling contributed with 2 uniquely detected bears and revealed the highest redundancy in detec-tion rates (Table 3). In 51 rub-tree sampling occasions, we ana-lyzed 2–4 samples from the same rub tree, detecting up to 2 bears 15.7% of the time.

Detection frequency over all sampling methods ranged from 1 to 11 sampling sessions per detected bear (X = ±3 2 2 7. . sessions), being higher for rub-tree sampling and lower for buckthorn and hair-trap sampling (Table 3). Overall, we detected 20 males and 25 females (1:1.25 MM:FF), but the observed sex ratio varied with sampling method (Table 3).

Three bear genotypes not known from previous surveys were judged to be sibling cubs, all sampled 3–5 times along with their putative mother at the same buckthorn and trap sites. This was corroborated by repeated sightings in the same area of a marked adult female with 3 cubs and further supported by the identification of a single male whose genotype could account for the nonmaternal alleles in each of the 3 putative siblings.

Population size and density.—Capture probabilities were influenced by sampling method, gender, exposure to noninva-sive sampling in previous surveys, and a variety of time- and effort-related covariates (Table 4). According to the most sup-ported models, hair-trap-capture probabilities ( p� = 0 15. ; 95% CI = 0.09–0.23) showed a linear trend, decreasing from p� = 0 18. (95% CI = 0.11–0.28) in session 1 to p� = 0 09.

(95% CI = 0.04–0.16) in session 5. Distance from the backbone of the study area (dfc) only marginally affected hair-trap-cap-ture probability (model 4), but its effect cannot be confidently confirmed (β = −0.244; 95% CI = −0.65–0.16). Capture prob-abilities at buckthorn sites ( p� = 0 22. ; 95% CI = 0.14–0.34) were strongly influenced by distance to the nearest sampling site (dbt; β = −4.593; 95% CI = −7.11–−2.08; Fig. 3) and were also affected by an interaction between temporal variation and the bears’ mean capture location (t × lat; Fig. 4). The similar AICc values of models 1 and 2 (Table 4) suggest a possible effect of gender on buckthorn capture probability, but this was not significant (Fig. 4). Rub-tree capture probabilities ( p� = 0 15. ; 95% CI = 0.11–0.20) varied by rub-tree effort (rte), number of

Table 3.—Descriptive statistics of 599 Apennine bear (Ursus arctos marsicanus) hair samples collected in the National Park of Abruzzo Lazio and Molise (June–September 2011) by means of 4 noninvasive sampling methods (HS: systematic hair-snagging; RT: rub-tree sampling; BT: sampling at buckthorn sites; INC: incidental sampling).

Total Sampling method

HS RT BT INC

Collected 679 159 278a 139 67 To lab 599 159 253b 122 65c

Replicates not analyzed 28 — 27 — 1 Inadequate 42 17 4 17 4Analyzed 529 142 222 105 60 Culled 103 40 40 6 17 Successful 426 102 182 99 43 % Successful 80.5 71.8 84.3 94.3 71.7No. bears detected 45 26 21 22 12Sampling redundancyd 9.5 3.9 8.7 4.5 3.6Males:females 1:1.25 1:1.17 1:1.10 1:1.75 1:0.71No. bears not previously knowne 16 8 4 8 2Uniquely detected bearsf — 8 2 8 2Detection frequencyg 3.2 ± 2.7 1.3 ± 0.5 3.2 ± 2.5 1.5 ± 0.6 1 ± 1Euro/genotypeh 706 328 635 287 300

a Including 43 alleged backup samples (i.e., replicated samples from the same rub-tree sampling occasion).b Including 33 alleged replicated samples.c Including 1 alleged replicated sample.d Number of successful samples by sampling method divided by number of detected genotypes.e Number of genotypes not known from previous noninvasive surveys (2000–2008) and livetrapping projects (2006–2010).f Number of bears uniquely sampled by a given sampling method.g Mean number of sessions (± SD) per detected bear.h Based on analyzed samples (n = 529) and accounting for costs of genetic analysis only (€60/analyzed sample).


rub trees available to bears (nrub), and the interaction between gender and bear detection during previous surveys (prev.hs). In particular, male bears which had been hair-snagged in previous surveys (i.e., aged > 4 years; see Appendix II) had

a higher per-session capture probability ( p� = 0 49. ; 95%

CI = 0.36–0.65) compared to never hair-snagged males (likely

≤ 4 years; p� = 0 10. ; 95% CI = 0.06–0.16; Fig. 5). We did not observe the same difference between previously ( p� = 0 16. ; 95% CI = 0.10–0.22) and never ( p� = 0 12. ; 95% CI = 0.09–0.17) hair-snagged females (Fig. 5). Mean capture probability for incidental sampling was p� = 0 23. (95% CI = 0.13–0.36), with no significant associations with the covariates we tested. Distance to the edge of the sampling grid for all sampling methods appeared to have a marginal effect (model 3), even though we cannot confidently confirm this (β = −0.152; 95% CI = −0.43–0.12).

Performing model averaging among all models, we estimated a superpopulation size of 51 bears (95% CI = 47–66 bears; CV = 7.9%), corresponding to 23 (95% CI = 21–31) males and 28 (95% CI = 26–35) females, with a sex ratio of 1:1.22 (95% CI = 1:1.13–1.24) MM:FF and a closure-corrected density of 39.7 (95% CI = 36.6–51.4) bears/1,000 km2. Estimates of popu-lation size from reduced sampling designs were all less con-sistent and more imprecise than that estimated from the fully combined data set (Table 5). When combining hair-snag and buckthorn data only, estimated population size and its CV were 9.8% and 125% higher, respectively, compared to the complete design, possibly reflecting underestimation of male capture probability (Table 5). Conversely, when combining hair-snag and rub-tree data only, population size was underestimated by 17.6% and its precision decreased by 44.3%, likely a result of a larger proportion of females being invisible to sampling (Table 5).

Table 4.—Model selection results for the Huggins closed-population estimation applied to the 2011 survey data of the Apennine brown bear (Ursus arctos marsicanus) population in the PNALM, Italy, using 4 noninvasive sampling methods. The 15 most-supported models, sorted by sample-size AICc values, are shown. Several other models, which received negligible support from the data, have been fitted (not listed).

Model Descriptiona K Deviance AICc ΔAICcb wic

1 HS(T) BT(t × lat + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 16 596.37 629.02 0 0.142 HS(T) BT(t × lat + dbt + sex) RT(rte + sex × prev.hs + nrub) INC(null) 17 594.31 629.05 0.03 0.143 HS(T) BT(t × lat + dbt) RT(rte + sex × prev.hs + nrub) INC(null) + dfe2 17 594.92 629.65 0.63 0.104 HS(T + dfc) BT(t × lat + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 17 595.03 629.76 0.74 0.095 HS(T) BT(t + lat + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 13 603.48 629.91 0.89 0.096 HS(T) BT(t × lat + dbt) RT(rte + sex × prev.hs + nrub) INC(prev.hs) 17 595.22 629.95 0.94 0.087 HS(t + sex + lt + prev.hs) BT(t × lat + sex + bte + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 23 583.44 630.77 1.75 0.058 HS(T + behav.) BT(t × lat + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 17 596.23 630.96 1.94 0.059 HS(t + sex + lt + prev.hs) BT(t + lat + sex + prev.hs + bte + dbt) RT(rte + sex × prev.hs + nrub) INC(null) 21 592.94 636.05 7.03 0.0010 HS(T) BT(t × lat + dbt) RT(rte + sex × prev.hs) INC(null) 15 606.38 636.90 7.94 0.0011 HS(T) BT(t × lat + dbt) RT(sex × prev.hs + nrub) INC(null) 15 609.92 640.49 11.47 0.0012 HS(T) BT(t × lat + dbt) RT(T + sex × prev.hs + nrub) INC(null) 16 608.73 641.38 12.36 0.0013 HS(T) BT(t × lat + dbt) RT(rte + sex + prev.hs + nrub) INC(null) 24 594.63 644.08 15.06 0.0014 HS(T) BT(t × lat + dbt) RT(t + sex + prev.hs + nrub) INC(null) 15 618.31 648.88 19.86 0.0015 HS(T) BT(t × lat + sex) RT(rte + sex × prev.hs + nrub) INC(null) 16 642.25 674.90 45.88 0.00

a Parameter definition: HS = hair-snagging; BT = sampling at buckthorn sites; RT = rub-tree sampling; INC = incidental sampling; t = time, 1 parameter per session; T = linear trends in capture probability; sex = gender effect; behav. = behavioral response after the 1st hair-snag detec-tion; bte = temporal variation of sampling effort at buckthorn sites by session (i.e., cumulative number of sampling days per session multiplied by total length of barbed wire in all installed buckthorn sites); dbt = distance from each bear’s mean capture location to the closest buckthorn sampling site; prev.hs = previous history of hair-snagging during the 2003–2008 surveys; lat = latitude of each bear’s mean capture location; rte = temporal variability of sampling effort at rub trees by sampling session (i.e., cumulative number of effective sampling days for each rub tree within each session); nrub = number of installed rub trees effectively available to sampled bears (see text); lt = previous livetrapping experi-ence; dfe = distance from each bear’s mean capture location to closest edge of sampling grid; dfc = distance from each bear’s mean capture location to the NW–SE backbone of the study area.b Difference in AICc values between the ith model and the model with the lowest AICc value.c Akaike weights.

Fig. 3.—Average (± 95% CI) buckthorn capture probability for Apennine brown bears (Ursus arctos marsicanus) as a function of the distance from the nearest sampled buckthorn site (PNALM ecosystem, Italy, 17 August–30 September 2011). Estimates of capture probability are from the model with no gender effect (Table 4, model 1), for ses-sion 1 and for the mean latitude of the study area.


DiscussionWe achieved good precision for a particularly small bear pop-ulation by integrating 4 noninvasive data sources in a formal DNA-based mark–recapture framework. In addition, compared to previous surveys of the same bear population, we used a more powerful set of markers, further reducing the potential for mismatching individuals. Our 2011 abundance estimate of 51 (95% CI = 47–66) bears compared to that obtained in 2008 (40 bears, 95% CI = 37–52 bears—Gervasi et al. 2012) provides evidence that the Apennine brown bear population has not been declining.

Sampling and capture probability.—Using multiple data sources entirely based on noninvasive genetic sampling, we produced a precise estimate (CV = 7.9%) of the small Apennine brown bear population. The approach we adopted in 2011 did not require handling of bears, nor the presence of previously marked individuals in the population, being there-fore more practical for the future monitoring of this as well as other small-sized bear populations. Provided at least one of the sampling methods ensures that all bears in the population are accessible to sampling (i.e., they have a nonzero capture

probability), other sampling methods could be adopted even if they suffer from minimal sample coverage (Boulanger et al. 2008; Kendall et al. 2008). By exploring progressively reduced sampling designs, we detected a lower number of bears, and no reduced combination of sampling methods produced estimates that were comparable in precision to our best estimate using the fully combined data set (Table 5). These empirical findings lead us to conclude that, to pursue a comparable level of preci-sion, an ideal sampling design for future surveys of this small bear population should include all 4 sampling methods, with hair-snagging as the primary sample and 2 or more secondary methods to effectively improve sample coverage.

Hair-snagging accounted for 57.8% of the total number of detected bears and 50% of those detected for the first time in 2011 and increased the minimum number of known bears by 40% above that detected by the other sampling methods (Table 3). The average hair-trap-capture probability we reported in 2011 was higher than in 2004 ( p� = 0 03. —Gervasi et al. 2008), likely as a result of improved field techniques. It was, however, lower than in 2008 ( p� = 0 23. ; 95% CI = 0.15–0.33—Gervasi

Fig. 4.—Session-specific capture probabilities at buckthorn sites for male and female Apennine brown bears (Ursus arctos marsicanus) in the a) southern and b) northern parts of the PNALM ecosystem, Italy (17 August–30 September 2011). Estimates of capture probability are provided for bears at 1 km from the nearest sampled buckthorn site and according to model 1 in Table 4. Buckthorn effort was the cumulative number of sampling days in each session multiplied by the cumulative length of the barbed wire at each buckthorn site. Error lines represent 95% CI.

Fig. 5.—Session-specific rub-tree capture probabilities for male and female Apennine brown bears (Ursus arctos marsicanus) which a) had been sampled in previous surveys or b) were never sampled before (PNALM ecosystem, Italy, 8 June–5 October 2011). Estimates of cap-ture probability are from the most-supported model (Table 4, model 1) for the average number of rub trees available to bears (nrub). Rub-tree effort was the cumulative number of sampling days for each rub tree summed across all rub trees within each session. Error lines represent 95% CI. Note that the scale on the y-axis differs between the panels.


et al. 2012), possibly due to a weaning reaction of bears to the lure in this small population. This would also explain the negative trend in sampling rate we reported across hair-snag-ging sessions, emphasizing the need for secondary lures (e.g., Kendall et al. 2008; Sawaya et al. 2012) in further hair-trapping applications. A similar decrease in hair-trap-capture probability across sessions was reported by Sawaya et al. (2012) for those sessions lacking a secondary lure. With the marginal exception of distance from the backbone of the study area (dfc), no other tested covariate significantly affected hair-trap-capture prob-ability, but we suspect that a low statistical power could have determined lack of covariate support (e.g., gender) previously known to affect hair-trap-capture probability (e.g., Boulanger et al. 2002; Kendall et al. 2008; Gervasi et al. 2012; Sawaya et al. 2012).

Although hair-trap-capture probability was not particularly high in our 2011 survey, by concurrently using secondary sampling methods, we markedly improved sample coverage and increased by 125% the minimum number of bears sam-pled relative to the 2008 survey. Whereas the sex ratio varied by sampling method (Table 3), the full data set reflected a sex ratio comparable to that derived from the abundance estimate. Rub-tree, buckthorn, and incidental sampling cumulatively improved detection frequencies for mark–recapture modeling. The increased sample size allowed us to effectively model het-erogeneity in capture probabilities of these sampling methods, thus reducing the risk of overestimating capture probability (see also below).

Although unevenly distributed, the density of installed rub trees in the PNALM was comparable to that in other noninva-sive bear surveys (Glacier National Park: 8–10 rub trees/100 km2—Kendall et al. 2008; North Continental Divide Ecosystem: 21 rub trees/100 km2—Kendall et al. 2009; Banff National Park: 14 rub trees/100 km2—Sawaya et al. 2012). Detection frequency at rub trees was comparatively high (Table 3), accounting for 46.7% of the bears cumulatively detected and contributing 10% of uniquely detected bears. The spatial and temporal component of rub-tree sampling had a marked effect on capture probability at rub trees (Table 4). Rub-tree capture probability was also higher for males than for females through-out the survey, consistent with previously reported noninvasive surveys in North America (Kendall et al. 2008; Sawaya et al. 2012). Additionally, in our bear population, we had the unique

opportunity to use previous hair-snag detections of individual bears as a proxy of 2 broad age classes (h.prev; Appendix II), indicating that adult (i.e., ≥ 4 years) male bears are those most frequently using rubs (Fig. 5A). In contrast to other rub-tree applications (Kendall et al. 2008; Sawaya et al. 2012), we did not detect a clear increase in rub-tree capture probability for females as the season progressed, and the highest capture prob-ability for both sexes during the last rub-tree session was prob-ably an artifact of the corresponding peak in sampling effort (Fig. 5). Provided appropriate covariates are used to effectively account for heterogeneous sampling effort and uneven sample coverage, we concur that rub-tree sampling is a practical way to increase sample coverage in surveying bear populations (Kendall et al. 2008; Stetz et al. 2010; Sawaya et al. 2012). Nevertheless, based on our results, rub-tree sampling could be expected to target almost entirely, although not exclusively, adult males, with a large proportion of the bear population remaining invisible to this sampling (Table 3). Reliance on this method in small bear populations would likely correspond to underestimated and imprecise population size estimates, as we observed with the sampling design based on rub-tree sam-pling alone (Table 5). By specifically targeting a proportion of the bear population with a relatively high capture probability (i.e., adult males), rub-tree sampling, proportionally more than other sampling methods, could be expected to produce a large number of redundant samples, contributing to high laboratory costs (Table 3). The efficiency of rub-tree sampling could be improved by collecting only 1 hair sample per rub tree and by subsampling the number of trees per sampling occasion (Sawaya et al. 2012).

Buckthorn sampling proved a very effective secondary sam-pling method, increasing by 40% the number of bears detected compared to the other sampling methods (Table 3). The rel-atively high capture probability we obtained at buckthorn patches in 2011, compared to our preliminary application in 2004 (Gervasi et al. 2008), enabled us to effectively model the associated sources of heterogeneity (Table 4). As expected from previous surveys (Gervasi et al. 2008), buckthorn capture prob-ability was influenced by the distance to the nearest buckthorn sampling site, approaching zero for distances ≥ 3 km (Fig. 3). Local variation in alpine buckthorn phenology along a south–north gradient most likely accounted for the decrease in buck-thorn capture probability as the ripening season progressed in

Table 5.—Minimum number of bears sampled and estimates of population size of the Apennine brown bear (Ursus arctos marsicanus) using noninvasive genetic sampling and progressively simplified sampling designs (PNALM ecosystem, June–September 2011).

Sampling design Minimum number of bears detected Population estimates

Total Females Males^N SE CV%

n % n % n %

All data sources 45 100 25 100 20 100 51 4.05 7.9Hair-snag + buckthorn 38 84 22 88 16 80 56 9.99 17.8Hair-snag + rub trees 35 78 19 76 16 80 42 4.80 11.4Buckthorn + rub trees 33 73 19 76 14 70 39 5.31 13.6Hair-snag only 26 58 15 60 11 55 46 12.59 27.4Buckthorn only 22 49 14 56 8 40 33 12.87 39.0Rub trees only 21 47 11 44 10 50 26 4.88 18.8


the south (Fig. 4A) and for the asynchronous increase in the north during the 3rd session (Fig. 4B). Accordingly, capture probability at buckthorn sites was null in the first and last ses-sions in both north and south PNALM, indicating that these sessions were conducted too early and too late, respectively, with respect to the ripening of buckthorn berries. Female bears had a higher capture probability at buckthorn sites than males (model 2, Table 4; Fig. 4), possibly reflecting their higher use of buckthorn berries than males. Due to the females’ smaller size and lower absolute energy requirements than adult males, they are less constrained by a more frugivorous diet (Welch et al. 1997; Rode and Robbins 2000; McLellan 2011) and are accordingly expected to make a more substantial use of this soft fruit during early hyperphagia. Whereas we lack evidence that cubs in the Apennine bear population are vulnerable to hair-snagging in the spring (Gervasi et al. 2012), we did capture cubs in the 2011 survey using double-stranded hair traps at buckthorn sites. Coincident capture of dependent offspring and their mothers may violate the assumption of independent cap-ture probabilities in mark–recapture models, but variable detec-tion of bears within family groups is thought to cause minimal bias to population estimates (Boulanger et al. 2004; Kendall et al. 2008; Gervasi et al. 2012).

Finally, at no extra field cost, incidental sampling increased the minimum number of known bears 10% above that detected by the other sampling methods combined, although it contrib-uted marginally to the overall detection frequency (Table 3). While incidental sampling has a minimal sample coverage, it has been successfully adopted in other surveys of bear popu-lations (e.g., De Barba et al. 2010; Sawaya et al. 2012) and has been shown to extend coverage to conditioned or habitu-ated bears living in close proximity to human settlements. In our case, 70.1% of incidental samples were collected during verification of alleged damage, allowing us to detect 10 bears, 2 of which were individually known to cause recurrent damage. Whereas 8 of these bears have been sampled ≤ 4 times each, the 2 known problematic bears were sampled 12 and 13 times each, confirming that property damage is disproportionally caused by relatively few bears (Ciucci and Boitani 2008).

Population abundance.—Although comparing population densities may be misleading due to differences among studies and the extent of the area surveyed (Smallwood and Schonewald 1998), the closure-corrected density we reported for the core Apennine brown bear population (all ages) lays in the upper tail of brown bear densities reported for interior areas (range: 1.6–40.9 bears/1,000 km2—Miller et al. 1997; Kendall et al. 2008; Proctor et al. 2010; Kindberg et al. 2011). This supports the contention that habitat productivity in the PNALM eco-system is currently adequate for bears (Ciucci et al. 2014) and underlines the role of the PNALM as a last, critical stronghold of brown bears in the Apennines. Although no formal estimates of carrying capacity are available for the PNALM (Ciucci and Boitani 2008), this result argues that any long-term conser-vation strategy of this remnant bear population should safely assume it will not grow much further within its core range. A more realistic approach should envision the expansion of the bear range across the central Apennines, provided appropriate

conservation actions are implemented both within and out-side protected areas (Boscagli 1999; Ciucci and Boitani 2008; Anonymous 2011). Contrasting the abundance of the bear pop-ulation in 2011 with that in 2008 (40 bears; 95% CI = 37–52 bears—Gervasi et al. 2012) results in a yearly rate of increase of λ = 1.084 based on point estimates alone. However, we urge caution in interpreting this result based on 2 considerations. First, although both the 2008 and 2011 abundance estimates were quite precise (CV = 6.8% and 7.9%, respectively), their confidence intervals overlap considerably (26.3% and 33.3% for the 2011 and 2008 estimates, respectively). Accordingly, using the Delta method to estimate the variance about the yearly rate of increase (Buckland et al. 1993) yields 95% CI for λ that ranges from 0.933 to 1.235. Second, to estimate the bear population size in 2008, we combined hair-snagging, livetrap-ping, and resighting data (Gervasi et al. 2012). In doing so, we assumed that any correlation between livetrapping and resight-ing data was effectively mediated by the rather high overall capture probability ( p� = 0 31. ; 95% CI = 0.22–0.44) and the contribution of the other data sources. Correlation among data sources was nevertheless expected to generate some minimal negative bias to population estimates and a slight negative bias to variance estimates (Gervasi et al. 2012). Due to the sampling methods adopted in 2008, it is also possible that we overesti-mated capture probability for male bears, leading to an under-estimate of their abundance. In this light, examination of the 2011 abundance estimates obtained with progressively reduced sampling designs to mimic the effect of reduced sample cover-age also calls for a cautious interpretation of the 2008–2011 alleged population growth, as 5 of 7 estimates from reduced data sets were comparable to, or lower than, the 2008 estimate (Table 5).

Marker selection and power.—The genetic variability we revealed using 13 microsatellites confirms that Apennine brown bears are particularly depleted of genetic variability (Lorenzini et al. 2004; Swenson et al. 2011). Only brown bear populations in some Alaskan islands (Paetkau et al. 1998), in the Pyrenees (Taberlet et al. 1997), and East Cantabrians (Pérez et al. 2009) attain comparably low or lower levels of genetic variability. Low genetic diversity is not only a factor endanger-ing small brown bear populations in western Europe (Swenson et al. 2011) but it also increases the probability of sampling more than 1 bear sharing the same multilocus genotype (i.e., 0MM-pairs—Paetkau 2003). We addressed this problem by empirically estimating how many 0MM-pairs might have been present in our data set with various sets of markers. We found that a core set of 9 microsatellites (HE = 0.44–0.58) used in previous studies (Gervasi 2008, 2012) provided for a small but nontrivial risk of false matches but that the addition of markers REN144A06 (HE = 0.61) and CXX20 (HE = 0.62) eliminated match probability as a practical concern. Thus, we recommend that future analyses of individual identity involving Apennine bears use 11 of the 13 microsatellites that we selected (i.e., all except G10X and MSUT-2, Table 2) plus gender. In addition, as was done in this study, marker G10P (HE = 0.22) can be used to modestly improve power when comparing genotypes between


labs, functionally extending the comparison between new and old data sets to 11 common markers (including gender).

Supporting InformationThe Supporting Information documents are linked to this manuscript and are available at Journal of Mammalogy online (jmammal.oxfordjournals.org). The materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supporting data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.Supporting Information S1.—Estimation of the probability of remaining undetected to all previous DNA sampling efforts for those Apennine brown bears (Ursus arctos marsicanus) alive in the study area at the beginning of the 2011 sampling. See Appendix II for details.

AcknowledgmentsThis survey coincided with the ex ante phase of action E3 funded by the European Commission Life+ NAT/IT/000160 project (“Life Arctos”). Cofunding agencies were the PNALM Authority, the National Forest Service, and La Sapienza University of Rome (Department of Biology and Biotechnologies). We are grateful to the PNALM wardens and the Forest Service personnel for their invaluable field support. T. Altea, R. Bucci, D. Gentile, A. Grassi, R. Latini, P. Leone, G. Palozzi, M. Posillico, E. Trella, V. Salvatori, L. Sammarone, and C. Sulli provided administrative and logistical support. We are also grateful to the many technicians and students who helped with field work and, in particular, M. Antonelli, O. Gallo, D. Gentile, M. Guerisoli, G. Grottolo Marasini, J. Mausbach, and F. Quattrociocchi provided invaluable assistance during hair-trap installment and hair collection. We greatly appreci-ated the comments and editing suggestions provided by 2 anon-ymous referees on a previous version of this manuscript.

Literature CitedAnonymous. 2011. Piano d’azione nazionale per la tutela dell’orso

bruno marsicano—PATOM. Quaderni Conservazione della Natura no. 37, Min. Ambiente – ISPRA, Roma, Italy [In Italian].

Boscagli, G. 1999. Status and management of the brown bear in Central Italy (Abruzzo). Pp. 81–84 in Bears. Status survey and con-servation action plan (C. Servheen, S. Herrero, and B. Peyton, eds.). IUCN/SSC Bear and Polar Bear Specialist Groups, IUCN, Gland, Switzerland.

Boulanger, J., K. C. Kendall, J. B. Stetz, D. A. Roon, L. P. Waits, and D. Paetkau. 2008. Use of multiple data sources to improve DNA-based mark–recapture population estimates of griz-zly bears. Ecological Applications 18:577–589.

Boulanger, J., and B. McLellan. 2001. Closure violation in DNA-based mark–recapture estimation of grizzly bear populations. Canadian Journal of Zoology 79:642–651.

Boulanger, J., B. N. McLellan, J. G. Woods, M. F. Proctor, and C. Strobek. 2004. Sampling design and bias in DNA-based cap-ture–mark–recapture population and density estimates of grizzly bears. Journal of Wildlife Management 68:457–469.

Boulanger, J., G. C. White, B. N. McLellan, J. Woods, M. Proctor, and S. Himmer. 2002. A meta-analysis of grizzly bear DNA mark-recapture projects in British Columbia, Canada. Ursus 13:137–152.

Buckland, S. T., D. R. Anderson, K. P. Burnham, and J. L. Laake. 1993. Distance sampling: estimating abundance of biological popu-lations. Chapman & Hall, London, United Kingdom.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and inference: a practical information theoretic approach. Springer, New York.

Chao, A., and R. M. Huggins. 2005. Modern closed-population cap-ture–recapture models. Pp. 58–87 in Handbook of capture-recapture analysis (S. C. Amstrup, T. L. McDonald, and B. F. J. Manly, eds.). Princeton University Press, Oxford, United Kingdom.

Ciucci, P., and Boitani, L. 2008. The Apennine brown bear: a critical review of its status and conservation problems. Ursus 19:130–145.

Ciucci, P., E. Tosoni, G. Di Domenico, F. Quattrociocchi, and L. Boitani. 2014. Seasonal and annual variation in the food habits of the remnant Apennine bear (Ursus arctos marsicanus) popula-tion, central Italy. Journal of Mammalogy 95:572–586.

Clevenger, A. P., and F. J. Purroy. 1996. Sign surveys for esti-mating trends of a remnant brown bear Ursus arctos population in northern Spain. Wildlife Biology 2:275–281.

Cubaynes, S., et al. 2010. Importance of accounting for detec-tion heterogeneity when estimating abundance: the case of French wolves. Conservation Biology 24:621–626.

De Barba, M., L. P. Waits, P. Genovesi, E. Randi, R. Chirichella, and E. Cetto. 2010. Comparing opportunistic and systematic sampling methods for non-invasive genetic monitoring of a small translocated brown bear population. Journal of Applied Ecology 47:172–181.

EEA (European Environment Agency). 2006. Corine land cover 2006 raster data. http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-3#tab-gis-data. Accessed 3 April 2014.

Ennis, S., and T. F. Gallagher. 1994. PCR-based sex determina-tion assay in cattle based on the bovine Amelogenin locus. Animal Genetics 25:425–427.

Falcucci, A., P. Ciucci, L. Maiorano, L. Gentile, and L. Boitani. 2009. Assessing habitat quality for conservation using an inte-grated occurrence–mortality model. Journal of Applied Ecology 46:600–609.

Gerber, B. D., J. S. Ivan, and K. P. Burnham. 2014. Estimating the abundance of rare and elusive carnivores from photographic-sampling data when the population size is very small. Population Ecology 56:463–470.

Gervasi, V., et al. 2008. A preliminary estimate of the Apennine brown bear population size based on hair-snag sampling and multiple data-source mark–recapture Huggins models. Ursus 19:103–121.

Gervasi, V., P. Ciucci, J. Boulanger, E. Randi, and L. Boitani. 2012. A multiple data source approach to improve abundance esti-mates of small populations: the brown bear in the Apennines, Italy. Biological Conservation 152:10–20.

Gervasi, V., P. Ciucci, F. Davoli, J. Boulanger, L. Boitani, and E. Randi. 2010. Addressing challenges in non invasive cap-ture–recapture based estimates of small populations: an Apennine brown bear pilot-study and meta-analysis. Conservation Genetics 11:2299–2310.

Green, G. I., and D. J. Mattson. 2003. Tree rubbing by Yellowstone grizzly bears Ursus arctos. Wildlife Biology 9:1–9.

Huggins, R. M. 1991. Some practical aspects of a conditional likeli-hood approach to capture experiments. Biometrics 47:725–732.

http://jmammal.oxfordjournals.org

http://jmamma.oxfordjournals.org/lookup/suppl/doi:10.1093/jmamma/gyu029/-/DC1

http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-3#tab-gis-data

http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-3#tab-gis-data


Ivan, J. S., G. C. White, and T. M. Shenk. 2013. Using auxiliary telemetry information to estimate animal density from capture-recapture data. Ecology 94:809–816.

Jackson, R. M., J. D. Roe, R. Wangchu, and D. O. Hunter. 2006. Estimating snow leopard population abundance using photogra-phy and capture–recapture techniques. Wildlife Society Bulletin 34:772–781.

Janecka, J. E., et al. 2008. Population monitoring of snow leopards using noninvasive collection of scat samples: a pilot study. Animal Conservation 11:401–411.

Kelly, J. M., J. Betsch, C. Wultsh, B. Mesa, and L. S. Mills. 2012. Noninvasive sampling for carnivores. Pp. 47–69 in Carnivore ecol-ogy and conservation. A handbook of techniques (L. Boitani and R. Powell, eds.). Oxford University Press, Oxford, United Kingdom.

Kendall, K. C., J. B. Stetz, J. Boulanger, A. C. Macleod, D. Paetkau, and G. C. White. 2009. Demography and genetic structure of a recovering grizzly bear population. Journal of Wildlife Management 73:3–17.

Kendall, K. C., J. B. Stetz, D. A. Roon, L. P. Waits, J. B. Boulanger, and D. Paetkau. 2008. Grizzly bear density in Glacier National Park, Montana. Journal of Wildlife Management 72:1693–1705.

Kendall, W. L. 1999. Robustness of closed capture–recapture methods to violations of the closure assumption. Ecology 80: 2517–2525.

Kindberg, J., J. E. Swenson, G. Ericsson, E. Bellemain, C. Miquel, and P. Taberlet. 2011. Estimating population size and trends of the Swedish brown bear Ursus arctos population. Wildlife Biology 17:114–123.

Long, R. A., P. MacKay, W. J. Zielinski, and J. C. Ray (eds.). 2008. Noninvasive survey methods for carnivores. Island Press, Washington, DC.

Lorenzini, R., M. Posillico, S. Lovari, and A. Petrella. 2004. Noninvasive genotyping of the endangered Apennine brown bear: a case study not to let one’s hair down. Animal Conservation 7:199–209.

Lynam, A. J., A. Rabinowitz, T. Myint, M. Maung, K. T. Latt, and S. H. T. Po. 2009. Estimating abundance with sparse data: tigers in northern Myanmar. Population Ecology 51:115–121.

McComb, B., B. Zuckerberg, D. Vesely, and C. Jordan. 2010. Monitoring animal populations and their habitat: a practitioners guide. CRC Press, Boca Raton, Florida.

McDonald, L. L. 2004. Sampling rare populations. Pp. 11–74 in Sampling rare or elusive species (W. L. Thompson, ed.). Island Press, Washington, DC.

McLellan, B. 2011. Implications of a high-energy and low-protein diet on the body composition, fitness, and competitive abilities of black (Ursus americanus) and grizzly (Ursus arctos) bears. Canadian Journal of Zoology 89:546–558.

Miller, S. D., et al. 1997. Brown and black bear density estima-tion in Alaska using radio telemetry and replicated mark–resight techniques. Wildlife Monographs. The Wildlife Society, Bethesda, Maryland 133.

Mills, L. S., J. J. Citta, K. P. Lair, M. K Schwartz, and D. A. Tallmon. 2000. Estimating animal abundance using noninva-sive DNA sampling: promise and pitfalls. Ecological Applications 10:283–294.

Mondol, S., K. U. Karanth, N. S. Kumar, A. M. Gopalaswamy, A. Andheria, and U. Ramakrishnan. 2009. Evaluation of non-invasive genetic sampling methods for estimating tiger population size. Biological Conservation 142:2350–2360.

Mowat, G., and C. Strobeck. 2000. Estimating population size of grizzly bears using hair capture, DNA profiling, and mark-recapture analysis. Journal of Wildlife Management 64:183–193.

Mulders, R., J. Boulanger, and D. Paetkau. 2007. Estimation of population size for wolverines Gulo gulo at Daring lake, Northwest Territories, using DNA based mark-recapture methods. Wildlife Biology 13:38–51.

Nichols, J. D., and B. K. Williams. 2006. Monitoring for conserva-tion. Trends in Ecology and Evolution 21:668–673.

O’Connell, A. F., J. D. Nichols, and U. K. Karanth (eds.). 2011. Camera traps in animal ecology: methods and analyses. Springer Publishing, Tokyo, Japan.

Paetkau, D. 2003. An empirical exploration of data quality in DNA-based population inventories. Molecular Ecology 12:1375–1387.

Paetkau, D., et al. 1998. Variation in genetic diversity across the range of North American brown bears. Conservation Biology 12:418–429.

Paetkau, D., and C. Strobeck. 1998. Ecological genetic studies of bears using microsatellite analysis. Ursus 10:299–306.

Pérez, T., et al. 2009. Non-invasive genetic study of the endangered Cantabrian brown bear (Ursus arctos). Conservation Genetics 10:291–301.

Proctor, M. F., et al. 2010. Ecological investigations of grizzly bears in Canada using DNA from hair, 1995–2005: a review of methods and progress. Ursus 21:169–188.

Proctor, M. F., B. N. McLellan, C. Strobeck, and R. Barclay. 2005. Genetic analysis reveals demographic fragmentation of griz-zly bears yielding vulnerably small populations. Proceedings of the Royal Society of London, B. Biological Sciences 272:2409–2416.

Raymond, M., and F. Rousset. 1995. GENEPOP (version 1.2): pop-ulation genetics software for exact tests and ecumenicism. Journal of Heredity 86:248–249.

Rode, K. D., and C. T. Robbins. 2000. Why bears consume mixed diets during fruit abundance. Canadian Journal of Zoology 78:1640–1645.

Roon, D. A., L. Waits, and K. C. Kendall. 2005. A simulation test of the effectiveness of several methods for error-checking non-inva-sive genetic data. Animal Conservation 8:203–215.

Ruell, E. W., S. P. D. Riley, M. R. Douglas, J. P. Pollinger, and K. R. Crooks. 2009. Estimating bobcat population sizes and densi-ties in a fragmented urban landscape using noninvasive capture–recapture sampling. Journal of Mammalogy 90:129–135.

Sawaya, M. A., J. B. Stetz, A. P. Clevenger, M. L. Gibeau, and S. T. Kalinowski. 2012. Estimating grizzly and black bear popula-tion abundance and trend in Banff National Park using noninvasive genetic sampling. PLoS One 7:e34777.

Schwartz, M. K., G. Luikart, and R. S. Waples. 2007. Genetic monitoring as a promising tool for conservation and management. Trends in Ecology and Evolution 22:25–33.

Smallwood, K. S., and C. Schonewald. 1998. Study design and interpretation of mammalian carnivore density estimates. Oecologia 113:474–491.

Sollmann, R., et al. 2013. Using multiple data sources provides density estimates for endangered Florida panther. Journal of Applied Ecology 50:961–968.

Stetz, J. B., K. C. Kendall, and C. Servheen. 2010. Evaluation of bear rub tree surveys to monitor grizzly bear population trends. Journal of Wildlife Management 74:860–870.

Swenson, J. E., P. Taberlet, and E. Bellemain. 2011. Genetics and conservation of European brown bears Ursus arctos. Mammal Review 41:87–98.

Taberlet, P., et al. 1997. Non-invasive genetic tracking of the endangered Pyrenean brown bear population. Molecular Ecology 6:869–876.

Teerink, B. J. 1991. Hair of West-European mammals: atlas and identifi-cation key. Cambridge University Press, Cambridge, United Kingdom.


Tosoni, E. 2010. Indagini ecologiche relative alla popolazione di orso bruno marsicano (Ursus arctos marsicanus) nel territorio del Parco Nazionale d’ Abruzzo, Lazio e Molise. Ph.D. dissertation, University of Rome “La Sapienza”, Rome, Italy [In Italian].

Waits, L. P., G. Luikart, and P. Taberlet. 2001. Estimating the probability of identity among genotypes in natural populations: cau-tions and guidelines. Molecular Ecology 10:249–256.

Waits, L. P., and D. Paetkau. 2005. Noninvasive genetic sam-pling tools for wildlife biologists: a review of applications and recommendations for accurate data collection. Journal of Wildlife Management 69:1419–1433.

Welch, C., J. Keay, K. Kendall, and C. T. Robbins. 1997. Constraints on frugivory by bears. Ecology 78:1105–1119.

White, G. C., and K. P. Burnham. 1999. Program MARK: sur-vival estimation from populations of marked animals. Bird Study Supplement 46:120–138.

White, G. C., K. P. Burnham, and D. R. Anderson. 2002. Advanced features of program MARK. Pp. 368–377 in Integrating people and wildlife for a sustainable future. Proceedings of the Second International Wildlife Management Congress, Gödölló, Hungary (R. Fields, R. J. Warren, H. Okarma, and P. R. Seivert, eds.). The Wildlife Society, Bethesda, Maryland.

White, G. C., and T. M. Shenk. 2001. Population estimation with radio marked animals. Pp. 329–350 in Design and analysis of radio telemetry studies (J. J. Millspaugh and J. M. Marzluff, eds.). Academic Press, San Diego, California.

Woods, J. G., D. Paetkau, D. Lewis, B. N. McLellan, M. Proctor, and C. Strobeck. 1999. Genetic tagging of free-ranging black and brown bears. Wildlife Society Bulletin 27:616–627.

Zedrosser, A., B. Dahle, J. E. Swenson, and N. Gerstl. 2001. Status and management of the brown bear in Europe. Ursus 12:9–20.

Submitted 17 January 2014. Accepted 17 July 2014.

Associate Editor was Marjorie Matocq.

Appendix IVariability of the Apennine bear (Ursus arctos marsicanus) at the additional 16 microsatellite markers which were not used for individual multilocus genotyping in this study, based on tis-sue and blood samples (≤ 25 bears). See Table 2 for details.

Appendix IIPrior to 2011, 3 extensive DNA surveys of the core Apennine brown bear population had been conducted in 2004, 2007, and 2008 (Gervasi et al. 2008, 2010, 2012), and no new genotypes were sampled during the 2007 and 2008 surveys. Hence, any bear that at the beginning of the 2011 survey had already been sampled in previous surveys was born before 2007. Based on each bear’s previous detection history, we therefore used a binary variable (prev.hs: previous hair-snag history) separating bears > 4 years old from younger bears (≤ 4 years) in order to account for the potential effect of a bear’s age on capture probability. Age classification is problematic for those bears sampled in 2011 but not before, as these comprise bears actu-ally born between 2007 and 2011 (i.e., ≤ 4 years old) and bears born before 2007 which survived through 2011 but remained undetected in previous surveys. Clearly, only if the latter prob-ability is reasonably small we can use prev.hs as a proxy of age class. To estimate this probability we used the formula:

P p p pa g g g g c aa

,*

, , ,{[( ) ( ) ( )]}= − ⋅ − ⋅ − ⋅ ⋅ −1 1 12004 2007 20081φ φ

where Pa g,* is the probability for an individual of age a and gen-

der g to remain invisible to sampling before 2011. Sex-specific capture probabilities in the 2004, 2007, and 2008 surveys were derived from Gervasi et al. (2008, 2010, 2012), whereas ϕc and ϕa are yearly survival probabilities for cubs and adults, respec-tively (P. Ciucci, pers. obs.), assuming these remain constant across survey years.

The complementary of Pa g,* (%corr) provides an estimate of the

proportion of bears correctly classified as ≤ 4 years, suggesting that the great majority of bears with no previous hair-snag history are expected to be ≤ 4 years old. This is especially true for male bears (%corr ≥ 0.96), whereas a certain risk of misclassification exists for females of the 2005–2007 cohorts (0.77 ≤ %corr ≤ 0.88), dropping to 8% (i.e., %corr = 0.92) for older females (Supporting Information S1).

Locus n HE HO A

G10Ha 25 0.46 0.52 2MSUT-6 23 0.35 0.35 2G1A 23 0.26 0.30 2D123 23 0.26 0.30 2CXX110 24 0.13 0.13 2G10U 23 0.04 0.04 2Mu15b 26 0.04 0.04 2CXX173 18 0.00 0.00 1Mu26 22 0.00 0.00 1G10O 23 0.00 0.00 1G10J 25 0.00 0.00 1REN145P07 23 0.00 0.00 1CPH9 22 0.00 0.00 1MU23 23 0.00 0.00 1G10M 23 0.00 0.00 1D1A 22 0.00 0.00 1

a This was the least variable of the 14 candidate microsatellites for individual identification based on the original marker selection data set (n = 25 bears; Table 2). As it is also comparatively long (> 250 bp) and likely to lower success rate with marginal samples, it was excluded.b Marker used by another lab in previous surveys of the Apennine brown bear population (Gervasi et al. 2008, 2012).



estimating abundance of the remnant apennine brown bear population using multiple noninvasive...

Documents